Enhancing Accuracy of Gujarati Word Tagging Using Advanced Learning Models

Gujarati, a morphologically rich and resource-poor Indian language, poses significant challenges for Natural Language Processing (NLP), particularly for Part-of-Speech (POS) tagging. Over the past two decades, research has evolved from rule-based and statistical models to hybrid systems and, more recently, deep learning and transformer-based approaches. This review paper systematically analyzes existing Gujarati POS tagging literature, taking reference from prior foundational and recent works, and presents a comparative discussion of methodologies, datasets, experiments, and results. In addition to synthesizing reported findings, this paper introduces new experimental evaluations using CRF, Bi-LSTM, and multilingual transformer models under a unified experimental setup. The results demonstrate clear performance gains with deep contextual models while highlighting trade-offs in computational cost and data requirements. The study concludes with research gaps and future directions for advancing Gujarati NLP.